An outlier mining algorithm based on constrained concept lattice

Traditional outlier mining methods identify outliers from a global point of view. These methods are inefficient to find locally biased data points (outliers) in low dimensional subspaces. Constrained concept lattices can be used as an effective formal tool for data analysis because constrained concept lattices have the characteristics of high constructing efficiency, practicability and pertinency. In this paper, we propose an outlier mining algorithm that treats the intent of any constrained concept lattice node as a subspace. We introduce sparsity and density coefficients to measure outliers in low dimensional subspaces. The intent of any constrained concept lattice node is regarded as a subspace, and sparsity subspaces are searched by traversing the constrained concept lattice according to a sparsity coefficient threshold. If the intent of any father node of the sparsity subspace is a density subspace according to a density coefficient threshold, then objects contained in the extent of the sparsity subspace node are considered as bias data points or outliers. Our experimental results show that the proposed algorithm performs very well for high red-shift spectral data sets.